> but we're looking at it technically. After 4 years of the
>core, they bring out another that's the same but with some
>added features.
Actually, Im quite in awe over those "some added features". The three most important ones, may well be among the best three ideas recently introduced into any MPU:
1) Hypertransport. glueless, cheap and fast SMP. the concept is really, really neat. I drooled all over those possible >4 way configurations showing different cpu's (yes, opterons and A64's) connected to each other, one cpu connected to the AGP tunnel, another to the PCI-X tunnel, each CPU having their own memory channel.. simple, but brilliant IMHO; makes designing SMP systems as easy as playing with lego
2) AMD64. I think we've discussed this already a bit
I'm a great fan of the concept.
3) integrated memory controller.
On the intel side, I'm a great admirer of hyperthreading. Also a beautiful technology, and pretty effective, at least for the P7. but other than that, there is little in the P7 core that I drool over. On the contrary, I think the most promising design from intel lately, is based on the more than 5 year old P6 core, ie banias, and soon Dothan. That really is the way forward IMHO, and I would love to see intel release a Dothan core that is slightly less optimized for low TDP, and gives us higher clockspeeds. At 2.4 GHz, Dothan could well be about the fastest cpu out there, bar none.
>and there were rumors a good while ago that K9 research had
>begun
Thats a given. A cpu design takes ~5 years easily; in fact I thought K8 and K9 where largely developped in parallel, and any idea's that could not be realized in time for K8, where handed to the K9 team.
>Now making sure the multicore works right and without
>problematic thermal control is gonna be a new challenge,
Easier than hyperthreading IMHO. K8 core is already multi core ready, it could be as "simple" as connecting both cores internally with hypertransport and adding cache coherency. The thermal density is another issue, but guestimates about opteron/A64 power consumption show its roughly half as hot as the upcomming prescott. So even without a die shrink, multicore should be feasable as a dual core will not use twice as much power as a single core, since a lot is shared (memory controller, HT, cache,..)
>and with SOI wafers costing already a lot, can we gamble at
>the potential cost PER chip containing 2 cores?
Actually, SOI wafers are cheaper than 300mm bulk wafers, and per CPU candidate, they are about even. Everyone always uses the 300mm wafers of intel as a cost argument, but right now, 200mm is cheaper per cpu. 300mm just helps getting more fab capacity until 300mm wafers come down in price.
>can we gamble at the potential cost PER chip containing 2
>cores?
If you take an opteron, roughly 60% of the die size is the 1 MB cache. The rest is for the memory controller, HT links and the core. So adding a second core while shrinking to a .09 process would roughly give you the same size of chip. IF AMD manages to gets in L2 caches a bit denser like intel, it should be a pretty small core really.
= The views stated herein are my personal views, and not necessarily the views of my wife. =